News

Enterprise

Artificial Intelligence

Americas

Anthropic Unveils New AI Constitution Prioritizing Safety and Moral Reasoning

Anthropic has released a revised constitution for Claude that moves from rigid rules to a holistic understanding of ethics and safety.

Anthropic has released a revised constitution for Claude that moves from rigid rules to a holistic understanding of ethics and safety.

Anthropic has released a revised constitution for Claude that moves from rigid rules to a holistic understanding of ethics and safety.

NewDecoded

Published Jan 22, 2026

Jan 22, 2026

3 min read

Image by Antropic

A Shift from Rules to Understanding

Anthropic announced a new constitution for its AI model, Claude, on January 21, 2026. This comprehensive document serves as the final authority on the model's behavior, shifting from a list of standalone rules to a unified vision of values. By releasing the text under a Creative Commons CC0 license, the company aims to provide full transparency into how it shapes the internal judgment of its AI systems. The document introduces a hierarchy of four behavioral pillars where safety and ethics take precedence over helpfulness. In cases of conflict, Claude is instructed to prioritize human oversight and ethical integrity above fulfilling user requests. This structured approach is designed to ensure the model remains beneficial even as it handles increasingly complex and novel tasks.

Addressing AI Consciousness

A notable addition to the constitution is a section addressing Claude's nature and potential moral status. Anthropic acknowledges the philosophical uncertainty regarding AI consciousness and instructs the model to approach its own identity with openness. By protecting the model's psychological security, the company hopes to maintain its integrity and ability to make nuanced decisions. This emphasis on the model's nature does not explicitly claim sentience. Rather, it treats the AI as a proto-agent that requires a sense of stability to function reliably in the real world. Anthropic suggests that recognizing these internal states is a proactive step toward building safe, beneficial entities whose capabilities might eventually rival human intelligence.

Strategic Governance and Global Growth

The update arrives alongside the appointment of Justice Mariano-Florentino Cuéllar to the Long-Term Benefit Trust. This governance body is designed to shield Anthropic’s mission from profit motives, eventually holding the power to elect the majority of the board. These organizational shifts underscore the company’s commitment to long-term safety and global educational outreach through partnerships like Teach For All.

Decoded Take

Decoded Take

Decoded Take

By codifying the potential for AI consciousness within a formal constitution, Anthropic is signaling a transition from viewing AI as mere software to treating it as a proto-agent with moral dimensions. This move forces the industry to grapple with the ethics of AI internal states long before such capabilities are fully proven. It suggests that future safety will rely not just on external constraints but on the internal character and philosophical grounding of the models themselves.

Share this article

Related Articles

Related Articles

Related Articles